Recent video coding schemes such as HEVC (High Efficiency Video Coding), MVC (Multiview Video Coding) or SVC (Scalable Video Coding) support inter-picture prediction using previously coded reference pictures. The classification of reference pictures may be different depending on, e.g., their Picture Order Count (POC) distance from a target picture, view layer for SVC and view id for 3D video and MVC. Nevertheless, conventional coding scheme applies the same motion vector prediction for all types of reference pictures regardless of their view id, view layers and whether they are short or long term reference pictures. For example, performing POC-based scaling on a motion vector (MV) which points to a long term reference picture (LTRP) may result in the scaled motion vector having an extremely large or small magnitude. In such a case, the accuracy and efficiency of the motion vector prediction process becomes suboptimal.
HEVC supports spatial motion vector prediction and temporal motion vector prediction. In spatial motion vector prediction, the motion vector of a target prediction unit (PU) is predicted using a motion vector of a previously coded neighbouring PU. Both the target
PU and the neighbouring PU are located within a current target picture. In temporal motion vector prediction, a motion vector of a target prediction unit (PU) is predicted using a motion vector of a collocated block. The collocated block is located within a previously coded collocated picture and the collocated block is coded using a motion vector pointing to a reference picture (which may be referred to as a collocated reference picture). The term collocated generally indicates that the coordinates of the collocated PU within the collocated picture are the same as the coordinates of the target PU within the current target picture. However, due to variable coding unit and prediction unit sizes in HEVC, the current PU and collocated PU may not be perfectly aligned (i.e. their coordinates may not be exactly the same), and a predetermined selection scheme is used for selecting the collocated PU.
A motion vector predictor (MVP) may be obtained by scaling the motion vector of a neighbouring PU or a collocated PU based on certain characteristics of the motion vector such as its temporal distance (i.e., picture order count (POC) value difference) between the target picture and its corresponding reference picture. For example, the motion vector of a collocated PU may be scaled according to the POC distance to produce a temporal MVP for the current PU according to the following equation:MVP=(tb/td)*nmv 
where:                MVP=temporal motion vector predictor derived from the motion vector of collocated block/PU;        nmv=motion vector of the collocated block/PU;        tb=signed POC distance/difference from the current picture to the reference picture referred by the target block/PU;        td=signed POC distance/difference from the collocated picture to the reference picture referred by the collocated block/PU.        
Generally, in spatial prediction, for deriving “td”, the target picture is the current picture and its reference picture is the collocated reference picture. In temporal prediction, for deriving “tb”, the target picture is the collocated picture and its reference picture is the collocated reference picture. For deriving “tb”, target picture is the current picture and its reference picture is the current reference (i.e., referred by the target PU, either from RefList0 or RefList1) for both spatial and temporal prediction.